Three Lines of Bash That Shook the AI World
A sheep farmer in Australia wrote three lines of bash. Eleven days later, OpenAI, Anthropic, and Hermes had all integrated his idea into their flagship AI coding tools.
This is the story of how Geoffrey Huntley, a shepherd who spends more time with sheep than with servers, accidentally reshaped the future of AI programming.
The Problem That Everyone Ignored
AI coding agents in 2026 are smart. Scary smart. They can write entire functions, debug complex systems, and refactor codebases in minutes.
But they have one fatal flaw: they never finish the job.
You give an agent a task. It runs three rounds, edits two files, then stops and asks: "What would you like me to do next?"
The bug isnt fixed. The feature isnt complete. But the AI is perfectly happy to move on.
Every major AI lab had this problem. None of them had solved it. Then a farmer stepped in.
The Ralph Loop
Geoffrey Huntleys solution was brutally simple: an infinite loop that keeps feeding the same prompt to Claude Code over and over. He called it the Ralph Loop, named after Ralph Wiggum from The Simpsons - that perpetually clueless but endlessly persistent kid who never gives up.
Raw. Inelegant. Devastatingly effective.
11 Days That Changed AI Coding
On April 30, OpenAIs Codex launched /goal. A week later, Hermes Agent followed. Then Claude Code. Eleven days. Three companies. Same command. Same inspiration.
Codex - Never Forgets: goals persist across restarts. One user ran a project for 14 hours, slept 5 hours, and Codex resumed exactly where it left off.
Hermes Agent - Never Quits: a multi-agent kanban system with five layers of protection: heartbeat detection, zombie reclamation, hallucination interception (verifying files actually exist), and retry budgets.
Claude Code - Doesnt Lie to Itself: after each round, a separate model (Haiku) judges if the job is actually done. The coder and the verifier are never the same entity.
Why This Matters
AI coding is shifting from generating code to closing the loop. The most important breakthrough of 2026 might not be a bigger, smarter model. It might be three lines of bash written by a man who spends his days herding sheep.